OCEANIC AND TERRESTRIAL SOURCES OF CONTINENTAL PRECIPITATION
Luis Gimeno,1 Andreas Stohl,2 Ricardo M. Trigo,3,4 Francina Dominguez,5 Kei Yoshimura,6 Lisan Yu,7 Anita Drumond,1 Ana María Durán-Quesada,1,8 and Raquel Nieto1 Received 18 January 2012; revised 31 August 2012; accepted 5 September 2012; published 8 November 2012.
[1] The most important sources of atmospheric moisture at Indonesia. Some landmasses only receive moisture from the global scale are herein identified, both oceanic and ter- the evaporation that occurs in the same hemisphere (e.g., restrial, and a characterization is made of how continental northern Europe and eastern North America), while others regions are influenced by water from different moisture receive moisture from both hemispheres with large seasonal source regions. The methods used to establish source-sink variations (e.g., northern South America). The monsoonal relationships of atmospheric water vapor are reviewed, and regimes in India, tropical Africa, and North America are the advantages and caveats associated with each technique provided with moisture from a large number of regions, are discussed. The methods described include analytical highlighting the complexities of the global patterns of and box models, numerical water vapor tracers, and physical precipitation. Some very important contributions are also water vapor tracers (isotopes). In particular, consideration is seen from relatively small areas of ocean, such as the given to the wide range of recently developed Lagrangian Mediterranean Basin (important for Europe and North techniques suitable both for evaluating the origin of water Africa) and the Red Sea, which provides water for a large that falls during extreme precipitation events and for estab- area between the Gulf of Guinea and Indochina (summer) lishing climatologies of moisture source-sink relationships. and between the African Great Lakes and Asia (winter). As far as oceanic sources are concerned, the important role The geographical regions of Eurasia, North and South of the subtropical northern Atlantic Ocean provides moisture America, and Africa, and also the internationally important for precipitation to the largest continental area, extending basins of the Mississippi, Amazon, Congo, and Yangtze from Mexico to parts of Eurasia, and even to the South Rivers, are also considered, as is the importance of terrestrial American continent during the Northern Hemisphere winter. sources in monsoonal regimes. The role of atmospheric In contrast, the influence of the southern Indian Ocean and rivers, and particularly their relationship with extreme events, North Pacific Ocean sources extends only over smaller con- is discussed. Droughts can be caused by the reduced supply tinental areas. The South Pacific and the Indian Ocean repre- of water vapor from oceanic moisture source regions. Some sent the principal source of moisture for both Australia and of the implications of climate change for the hydrological cycle are also reviewed, including changes in water vapor concentrations, precipitation, soil moisture, and aridity. It is important to achieve a combined diagnosis of moisture sources using all available information, including stable 1Ephyslab, Departamento de Física Aplicada, Facultad de Ciencias de Ourense, Universidad de Vigo, Ourense, Spain. water isotope measurements. A summary is given of the 2NILU - Norwegian Institute for Air Research, Kjeller, Norway. major research questions that remain unanswered, including 3CGUL, IDL, University of Lisbon, Lisbon, Portugal. (1) the lack of a full understanding of how moisture sources 4Departamento de Engenharias, Universidade Lusófona, Lisbon, influence precipitation isotopes; (2) the stationarity of Portugal. moisture sources over long periods; (3) the way in which 5Department of Atmospheric Sciences, University of Arizona, Tucson, Arizona, USA. possible changes in intensity (where evaporation exceeds 6Atmosphere and Ocean Research Institute, University of Tokyo, precipitation to a greater of lesser degree), and the loca- Tokyo, Japan. tions of the sources, (could) affect the distribution of con- 7 Department of Physical Oceanography, Woods Hole Oceanographic tinental precipitation in a changing climate; and (4) the Institution, Woods Hole, Massachusetts, USA. 8Now at Department of Atmospheric, Oceanic and Planetary Physics role played by the main modes of climate variability, such (School of Physics) and the Center for Geophysical Research, University as the North Atlantic Oscillation or the El Niño–Southern of Costa Rica, San Jose, Costa Rica. Oscillation, in the variability of the moisture source regions, as well as a full evaluation of the moisture transported by low-level jets and atmospheric rivers. Corresponding author: L. Gimeno, Ephyslab, Departamento de Física Aplicada, Facultad de Ciencias de Ourense, Universidad de Vigo, Campus As Lagoas s/n, ES-32004 Ourense, Spain. ([email protected])
©2012. American Geophysical Union. All Rights Reserved. Reviews of Geophysics, 50, RG4003 / 2012 1of41 8755-1209/12/2012RG000389 Paper number 2012RG000389 RG4003 RG4003 GIMENO ET AL.: SOURCES OF CONTINENTAL PRECIPITATION RG4003
Citation: Gimeno, L., A. Stohl, R. M. Trigo, F. Dominguez, K. Yoshimura, L. Yu, A. Drumond, A. M. Durán-Quesada, and R. Nieto (2012), Oceanic and terrestrial sources of continental precipitation, Rev. Geophys., 50, RG4003, doi:10.1029/2012RG000389.
1. INTRODUCTION hydrological cycle [Trenberth et al., 2011]. There has also been a dramatic increase in the number of water vapor iso- [2] Given the importance of global climate change, an understanding of the nature and intensity of the hydrological topes observations [Risi et al., 2012], which are fundamental cycle and of its development over time is one of the most to the validation of analytical and numerical models [e.g., pressing challenges currently faced by mankind. Although Yoshimura et al., 2004]. Global circulation models with the atmosphere contains only a small proportion of the total advanced cloud microphysics and a realistic representation of global water, it nevertheless plays a key role in connecting orography have also incorporated new parametrizations that the major reservoirs of the oceans, lakes, soils, inland and better represent processes involving soil moisture and have sea ice, and rivers via the transport of moisture, evapo- afforded significant improvements to the ability of general transpiration, and precipitation. Water vapor accounts for circulation models (GCMs) to represent the atmospheric only about 0.25% of the total mass of the atmosphere water cycle [Andersson et al., 2005]. Furthermore, the “ ” [Seidel, 2002], but its importance in regulating global cli- trajectory-based ( Lagrangian ) methods used to identify mate and weather patterns is beyond dispute [Held and sources of moisture available for precipitation have been Soden, 2000]. The hydrological cycle may be summarized widely used to assess both global [e.g., Stohl and James, as the evaporation of moisture at one location and precipi- 2005; Dirmeyer and Brubaker, 2007; Gimeno et al., 2010a] tation elsewhere, balanced by the atmospheric, oceanic, and and regional sources [e.g., Nieto et al., 2006; Sodemann hydrological transport of water. In oceanic regions, the rate et al., 2008]. [4] In the following sections, recent work related to all the of evaporation generally exceeds the rate of precipitation, foregoing different aspects of the hydrological cycle is and oceans therefore represent a net source of moisture that summarized, but with a focus on the atmospheric part of the is then transported by the atmosphere to the continents; hydrological cycle. The review concentrates on works pub- landmasses act as net sinks of atmospheric moisture where lished in the last three decades, but there is more historical precipitation exceeds evapotranspiration. Surface water then feeds rivers, groundwater, and other bodies that discharge information that is not being discussed here. In Section 2, the general distribution of evaporation, water vapor, and pre- into the ocean, thereby completing the cycle. In global cipitation is described, as are the general patterns of water terms, the hydrological cycle is responsible for an annual vapor transport. In Section 3, the source-sink relationships rate of evaporation of about half a million cubic kilometers are examined, first in a discussion of the different methods, of water, around 86% of which is from the oceans, with the their assumptions, and their advantages and disadvantages, remainder having its origin in the continents [Quante and and second by summarizing the main evaporative source Matthias, 2006]. Most of the water that evaporates from regions and transport paths of moisture for global and the oceans (90%) is precipitated back into them. Only 10% falls as precipitation over the continents (Figure 1). Of this regional precipitation. In Section 4, the transport of moisture during extreme episodes such as drought and flood events is precipitation, approximately two thirds is recycled over the discussed. In Section 5, some of the implications of climate continents, and only one third runs off directly into the change for the hydrological cycle are reviewed, and it is oceans [e.g., Trenberth et al., 2007a]. Because human society proposed that if it is indeed critical to understand the pro- is becoming increasingly reliant on the security of its fresh- cesses that govern moisture transport in the troposphere, it water resources, and has adapted to the present-day hydro- is even more so in a changing climate [Christensen and logical cycle and in particular to the current precipitation Christensen, 2003; Schär et al., 2004]. To understand the regime, it is essential to understand the processes of evapo- ration from the oceans (via the study of oceanography [Yu, transport is to understand the relationship among the changes in evaporation, in atmospheric moisture content, and in pre- 2007]), the transport of atmospheric moisture (meteorology cipitation, which provides the only means of explaining why [Trenberth et al., 2003]), and the effects of these two pro- the patterns predicted by different climate models differ so cesses in particular on the hydrological cycle (hydrology substantially. In the final section (Section 6), some topics are [Bales, 2003]), all of which are affected by global climate highlighted that require further research in the coming years. change [Intergovernmental Panel on Climate Change (IPCC), 2007]. [3] Recent years have seen an increasing number of studies 2. GLOBAL DISTRIBUTION OF WATER VAPOR using novel remote sensing techniques, which has allowed 2.1. Evaporation and Precipitation ever more sophisticated and robust estimates of oceanic [5] Evaporation is the process by which water molecules evaporation to be made (e.g., the Objectively Analyzed air- change phase from liquid to gas. Turbulent eddies transport sea Flux project (OAFlux) [Yu et al., 2008]). New data moisture away from the evaporating surface. For practical assimilation methods have improved meteorological reana- applications, we simplify these turbulent fluxes using bulk lyses, which now provide a much better closure of the
2of41 RG4003 GIMENO ET AL.: SOURCES OF CONTINENTAL PRECIPITATION RG4003
Figure 1. The hydrological cycle. Estimates of the observed main water reservoirs (black numbers, in 103 km3) and the flow of moisture through the system (red numbers, in 103 km3 yr 1). Adjusted from Trenberth et al. [2007a] for the period 2002–2008 as in Trenberth et al. [2011]. transfer coefficients to relate the fluxes to the mean properties reanalysis outputs [e.g., Large and Yeager, 2009]. One such of the flow. Consequently, evaporation E can be expressed as product was developed by the OAFlux project [Yu and Weller, 2007; Yu et al., 2008]. Figures 2a and 2b show the ¼ ¼ ðÞ ; ð Þ E ceUdq ceUqs qa 1 temporally averaged ocean evaporation for January and July. Oceanic evaporation obtained from other data sets is quali- where U is the near-surface wind speed, ce is a turbulent exchange coefficient, q is the saturation specific humidity at tatively similar in terms of its main characteristics, although s significant quantitative differences exist [e.g., Andersson the evaporating surface, and qa is the near-surface atmo- spheric specific humidity. This basic equation is then modi- et al., 2011]. [7] Over land, equation (1) is usually presented in a fied to reflect the nature of the evaporating surface. Over the slightly different form, using bulk aerodynamic resistance oceans, the following parametrization [Fairall et al., 2003] is (ra) rather than the turbulent exchange coefficient (ce), where often used: 1 ra =(ce U ) , using vapor pressures rather than specific ≈ E ¼ ceUdq ¼ ceUqðÞsðÞ SST qaðÞTa; RH ; ð2Þ humidity, and assuming q 0.622e/p: where q is the saturation specific humidity for a given sea 0:622rðÞesðÞ T0 eTðÞ s E ¼ ; ð3Þ surface temperature (SST) and qa is the near-surface atmo- psra spheric specific humidity. [6] The global distribution over ocean of E is commonly where the constant 0.622 is the ratio of the molecular weight constructed from equation (2) using air-sea variables that can of water vapor to the effective molecular weight of dry air, be obtained from satellite observations [e.g., Chou et al., es(T0) is the saturation vapor pressure for a given surface 2003; Kubota and Tomita, 2007; Andersson et al., 2011] temperature T0, e is the vapor pressure above the surface, T is and/or from reanalysis data. A key limitation of satellite data the near-surface air temperature, and ps is the atmospheric is the challenge of retrieving near-surface air humidity and pressure at the surface. Meteorological observations over temperature [e.g., Curry et al., 2004; Yu, 2009], which land generally only provide the temperature 2m above the requires certain assumptions to be made. To reduce this surface, and for this reason the Penman-Monteith equation problem, satellite observations were combined with may be derived from equation (3) and the expression for
3of41 RG4003 GIMENO ET AL.: SOURCES OF CONTINENTAL PRECIPITATION RG4003
Figure 2. Ocean time-mean rates of (a, b) E, (c, d) P, and (e, f) E-P for January and July. E is from OAFlux [Yu and Weller, 2007] for 1988–2008, P is from GPCP [Adler et al., 2003] for 1988–2008, and E-P is the combination of these. sensible heat flux (see Shuttleworth [2012] for a derivation) [9] A complete review of the basic theories, observational in order to give an expression for evaporation that only methods, satellite algorithms, and land surface models for requires observations of humidity, temperature and wind evaporation over land may be found in Wang and Dickinson speed at a single level: [2012]. The principal methods of measuring evapotranspi- ration are summarized in Table 1 (eddy covariance, Bowen rcp DðÞþRn ðÞesðÞ T eTðÞ ratio (BR), weighable lysimeters, scintillometer, surface ¼ ra ; ð Þ LvE 4 water balance, and atmosphere water balance methods), as rs D þ g 1 þ reviewed by the authors. ra [10] Figure 3 shows both the ensemble average and the where L is the latent heat of vaporization, D is the slope of uncertainty of the mean annual and seasonal values of global v – the saturation vapor pressure versus temperature curve at evapotranspiration for the period 1984 2007, as derived using two surface radiation budget products and three process- temperature T, Rn is the net incoming radiation, r is the based models [from Vinukollu et al., 2011]. The ensemble density of air, cp is the specific heat of air, g = cp p/(0.622 Lv), mean shows the spatial distribution of evapotranspiration, and rs is the canopy-averaged leaf stomatal resistance obtained using the big-leaf approximation [see Shuttleworth, with low values in arid regions, highest values in the humid 2012]. The Penman-Monteith equation (4) is perhaps the best tropics, and intermediate values in midlatitude forests and known expression for evaporation over land. agricultural regions. The seasonal cycle shows the greening [8] Over land, the global network of eddy covariance (EC) of the midlatitudes during their respective hemispheric spring towers (towers that measure surface fluxes based on turbu- and summer. There is some interseasonal variability in the lence theory) FLUXNET provides continuous data on water uncertainties, which are greatest in humid tropical and sub- and energy fluxes for a wide range of ecosystems and cli- tropical monsoon regions. mates [Baldocchi et al., 2001]. At a larger scale, recent [11] Once evaporated, water vapor molecules typically merged flux tower and satellite data [Reichstein et al., 2007; spend about 10 days in the atmosphere before condensing Mu et al., 2007] and merged satellite and gridded climate and falling to the Earth as precipitation [Numaguti, 1999]. data [Fisher et al., 2008] provide global estimates of ter- The 10 day period considered is a median of a broad prob- restrial evapotranspiration. ability density function of residence times of water vapor in the atmosphere. Most of the water vapor evaporated from the
4of41 RG4003 GIMENO ET AL.: SOURCES OF CONTINENTAL PRECIPITATION RG4003
TABLE 1. A Summary of Observation and Estimation Methods for Evapotranspirationa
Method Temporal Scale Spatial Scale Advantages Disadvantages Eddy Covariance Half hour to yearly. Hundreds of m depending Direct measurement of Regional and global estimation on measurement height turbulence fluxes and can be made. above canopy layer and independent observation. wind speed. Bowen Ratio Half hour to yearly. Hundreds of m depending Energy is balanced. Diffusivity for water on measurement height and heat are assumed above canopy layer to be equal. Energy and wind speed. balance is assumed (energy components are point measurements and fluxes have a large footprint). Lysimeter Half hour to yearly. Point measurement. Direct observation. Environment is disturbed. Scintillometer Half hour to yearly. Tens of m to tens Captures turbulence Depends on MOST of km. fluxes over large universal functions. scale with known footprints. Surface Water Balance Monthly to yearly. Hundreds to thousands Direct estimate, Accuracy can only of km. regional and be guaranteed at global estimation low temporal can be made. (multiyear average) and spatial resolution. Atmospheric Water Balance Monthly to yearly. Hundreds to thousands Regional and global Low accuracy. of km. estimation can be made. aFrom Wang and Dickinson [2012]. oceans falls back into the oceans as precipitation, while Liepert and Previdi, 2009], explanation of observed changes about 10% is transported over land and influences terrestrial in ocean salinity [Lagerloef et al., 2010; Bingham et al., hydrological processes [Oki, 2005]. The climatological 2010; Ren and Riser, 2009; Yu, 2011], estimation of the mean distribution of global precipitation rate, P, is shown in freshwater budget balance in regional and global oceans Figures 2c and 2d for January and July using the precipita- [Sanchez-Gomez et al., 2011; Schanze et al., 2010], and tion data set from the Global Precipitation Climatology inference of the mean and variability of the continental Project (GPCP [Huffman et al., 1997; Adler et al., 2003]). freshwater discharge to the global oceans [Seo et al., 2009; Other commonly used precipitation data sets include the Syed et al., 2010]. The balance of E and P indicates the major Tropical Rainfall Measuring Mission (TRMM) Multisatellite sources and sinks of water vapor over the globe. The major Precipitation Analysis (TMPA [Huffman et al., 2007]), the net sources (E > P) are located over the subtropical belts of Climate Prediction Center (CPC) Merged Analysis of high evaporation, and the major net sinks (E < P) are found Precipitation (CMAP [Xie and Arkin, 1997]), the precipita- in the Intertropical Convergence Zone (ITCZ), the South tion estimates from the CPC MORPHing technique Pacific Convergence Zone (SPCZ), and the midlatitude (CMORPH [Joyce et al., 2004]), the Unified Microwave storm tracks where the convection of moisture results in Ocean Retrieval Algorithm (UMORA [Hilburn and Wentz, high precipitation. 2008]), and Precipitation Estimation from Remotely Sensed 2.2. Water Vapor Flux and Divergence Information using Artificial Neural Networks (PERSIANN [Hsu et al., 1997]). [14] To gain improved understanding of the transport of [12] A combination of satellite-derived E and P data sets atmospheric moisture, great efforts have been made to yields estimates of global ocean freshwater flux. However, as advance space and in situ observational platforms to better pointed out by Schlosser and Houser [2007], these estimates quantify the distribution and variation of water vapor in the are quite uncertain because each time series is calibrated atmosphere. For instance, Ross and Elliott [1996] provided differently, data sources are usually inhomogeneous, and more quality-controlled long-term radiosonde observations in the critically, there are no comprehensive in situ validation data. United States, and these observations were later extended to [13] Nevertheless, in their study of the ocean freshwater the whole of the Northern Hemisphere [Ross and Elliott, budget (E-P) using ocean salinity observations Schanze et al. 2001]. Satellite observations have also been available for [2010] showed that among a variety of possibilities, the E-P some time thanks to Meteosat-3 and -4 [Pierrehumbert and pair from OAFlux E and GPCP P [Yu et al., 2008; Adler Roca, 1998], Special Sensor Microwave/Imager (SSM/I) et al., 2003] was the only pair capable of balancing the [Wentz and Schabel, 2000; Santer et al., 2007; Wentz et al., ocean freshwater budget within the measurement uncertain- 2007], High-Resolution Infrared Radiation Sounder (HIRS) ties (Figures 2e and 2f). The combined use of these two data [Bates et al., 2001], the Global Ozone Monitoring Experi- sets may be seen in a variety of applications, including the ment (GOME) [Wagner et al., 2005], Atmospheric Infrared validation of climate model simulations [e.g., Allan, 2009; Sounder (AIRS) [Dressler et al., 2008], Global Positioning
5of41 RG4003 GIMENO ET AL.: SOURCES OF CONTINENTAL PRECIPITATION RG4003
Figure 3. Map of (left column) the ensemble average, (middle column) ensemble range, and (right column) normalized ensemble range in global evapotranspiration for (top row) annual mean and (bottom four rows) seasonal means. The ensemble used outputs from two surface radiation budgets and three process-based evapotranspiration models. The normalized ensemble range is calculated as the range divided by the ensemble mean. From Vinukollu et al. [2011].
System (GPS) [Wolfe and Gutman, 2000], and other higher latitudes. If the total water vapor content in the techniques. atmosphere were to condense and precipitate, the depth of [15] The global distribution of water vapor is shown in precipitation would be about 50 mm at equatorial latitudes, Figures 4a and 4b for January and July using the total column but only about 5 mm at the poles [Quante and Matthias, water vapor (TCWV) obtained from SSM/I observations. As 2006]. The highest TCWV occurs over the tropical Pacific shown in Trenberth et al. [2011], the overall patterns and warm pool, and its location and seasonal variation are shown temporal variation of water vapor over the oceans generally in Figures 4a and 4b. follow those of SST, because according to the Clausius- [16] The global distribution of evaporation (Figures 2a and Clapeyron (C-C) equation, the saturation water vapor pres- 2b) differs from that of atmospheric water vapor (Figures 4a sure is a nonlinear function of temperature. According to C-C and 4b), and also from that of precipitation (Figures 2c and equation a change in temperature of 1 typically causes a 7% 2d). This is because for precipitation to occur, three factors change in water vapor content [Held and Soden, 2000; are important, namely (1) the availability of atmospheric Wentz et al., 2007]. Because of its sensitivity to temperature, moisture, (2) a cooling mechanism, and (3) the presence of the water vapor content is high in the lower atmosphere, cloud condensation nuclei (CCN). All of these are necessary and decreases with height. Moreover, water vapor occurs at for the condensation process to occur and for droplets to high concentrations in the tropics and is less prevalent at form and grow sufficiently large to fall out of the
6of41 RG4003 GIMENO ET AL.: SOURCES OF CONTINENTAL PRECIPITATION RG4003
Figure 4. Mean total column water vapor (TCWV) for (a) January and (b) July. Adapted from Trenberth et al. [2011]. atmosphere. Typically, cooling is caused by the uplift of an Equation (6) states thatZ the temporal rate of change of pre- air mass, either due to convection, large-scale ascent, or flow 1 ps cipitable water, W ¼ qdp, and the divergence of the over a topographic obstacle, but radiational cooling is also g 0 possible (e.g., through the formation of fog). Usually, con- water vapor transport integrated over the depth of the atmo- densation in the free atmosphere is not possible without the sphere (r Q) must balance the fresh water flux E-P at the presence of aerosols. It is the microphysics that controls the surface. formation of cloud droplets or ice crystals through collision [18] Early studies [e.g., Benton and Estoque, 1954; Starr or coalescence, as well as their growth and precipitation and Peixoto, 1958; Rasmusson, 1967] have demonstrated [Houze, 1993]. The global distribution of precipitation is that, provided that the water vapor flux Q can be measured more similar to the distribution of TCWV, particularly in the with sufficient accuracy, equation (6) is useful for evaluating tropics, in areas of low-level convergence and high SST. In the combined change in surface and subsurface water stor- the tropics, there is also far more structure to the patterns of age. Following these earlier publications, continuing efforts rainfall, due to the effects of major circulation regimes such have been made to estimate Q using available observational as the monsoons and the Hadley cell. data, such as those obtained from rawinsondes [e.g., [17] The transport of water vapor in the atmosphere is Rasmusson, 1967; Peixoto et al., 1981] and satellites [Liu typically represented by the vertically integrated total hori- and Tang, 2005; Xie et al., 2008], and also from atmo- zontal flux of water vapor, which can be expressed as spheric reanalyses [e.g., Trenberth and Guillemot, 1995; Mo Z and Higgins, 1996]. Satellite observations with near-global 1 ps Q ¼ qVdp; ð5Þ coverage and fine temporal and spatial resolution have g 0 shown great promise in improving the estimation of Q. where g is the acceleration due to gravity, p is the pressure, Figures 5a and 5b show the satellite-derived mean vector Q p is the pressure at the surface, q is the specific humidity, field of superimposed on the mean flux divergence s r Q Q and V is the horizontal wind vector at a given level, com- ( ) for January and July. The fields are constructed posed of both mean and eddy components. Using the con- from the combined use of multiple satellite observations, servation of mass, the hydrological balance in the atmosphere including near-surface wind vectors from QuikScatterometer can be formulated as follows: (QuikSCAT), cloud drift wind vectors from the Multi- angle Imaging Spectroradiometer (MISR) and geostationary ∂W þr Q ¼ E P; ð6Þ satellites, and precipitable water from SSM/I [Xie et al., ∂t 2008]. The transport of moisture integrated over the depth
Figure 5. Vector field of the vertically integrated total horizontal flux of water vapor Q (unit: kg/m/s) superimposed on the flux divergence (r Q; unit: cm/yr) for (a) January and (b) July. Data are from Xie et al. [2008] for 1999–2008.
7of41 RG4003 GIMENO ET AL.: SOURCES OF CONTINENTAL PRECIPITATION RG4003 of the atmosphere estimated over oceans using satellite data objective climatology, Knippertz and Wernli [2010] showed was validated using independent daily rawinsonde observa- that such exports of tropical moisture are most frequent in tions (a total of 28,408 rawinsonde observations), monthly four particular regions of the Northern Hemisphere, namely mean reanalysis data, and regional water balance [Xie et al., (1) the “Pineapple Express,” which connects tropical mois- 2008]. The means (standard deviations) of the differences ture sources near Hawai‘i with precipitation near the North between the two values of Q obtained from rawinsonde and American West Coast and has a marked peak in activity in satellite data were 2.75 kg/m/s (69.83) for DQx, and boreal winter; (2) over the western Pacific in summer; 8.58 kg/m/s (60.16) for DQy. The correlation coefficients (3) over the Great Plains of North America, starting over the between Q from rawinsonde and Q from satellite were 0.948 Gulf of Mexico and the Caribbean Sea and peaking in sum- for DQx, and 0.867 for DQy. By comparing time series mer and spring; and (4) over the western North Atlantic, with at individual rawinsonde stations it is seen that the satellite a maximum in winter and fall. Some of these ARs (like the data capture not only the seasonal changes but also the example shown in Figure 6) cause extreme precipitation and synoptic variations of the observations. Values of Q from floodings over those regions (e.g., the 1993 and 2008 floods the National Centers for Environmental Prediction (NCEP) over the central United States [Dirmeyer and Kinter, 2009], reanalysis data furthermore showed significant correlation flooding in western Washington [Neiman et al., 2011], in (with a correlation coefficient greater than 0.9 in most areas) California [Ralph and Dettinger, 2011], in the UK [Lavers with Q from satellite data over global oceans. et al., 2011], and in Norway [Stohl et al., 2008]). [19] There is a good agreement between the geographical 2.4. Limitations of Available Data Sets distributions of r Q in Figures 5a and 5b and E-P in and Uncertainties in the Estimation Figures 2e and 2f, demonstrating that, averaged over time, of the Components of the Water Budget the rate of change of water storage is small, and E-P is largely balanced by r Q. Throughout the year, the trans- [22] Over continental regions, a high density of precipita- port of water vapor in the tropics is characterized by a broad tion data is available, including for most of Europe, the band of easterly transport in the Atlantic Ocean and the United States, Australia and some parts of Asia. For large central and eastern Pacific and by a seasonal reversal of parts of Africa, continental South America, and some regions direction in the Indian Ocean and its vicinity, in association in Asia and northern North America, however, data are more with monsoons [Peixoto and Oort, 1992]. Outside the tro- scarce [New et al., 2001]. Prior to the advent of satellites, pics, water vapor is transported poleward. over the oceans all data were collected using shipborne in situ measurements. The ability of radiosondes to measure water 2.3. Long-Range Transport of Water Vapor vapor accurately has improved over time [Dai et al., 2011], [20] As shown in Figure 5, the strong easterly fluxes of although gaps in coverage and missing data remain problems moisture in the tropics are due to the highest global values of to be overcome. precipitable water. Almost equally strong fluxes occur [23] Schanze et al. [2010] reviewed the temporal evolution around the major subtropical anticyclones in the summer of the availability of data (Table 2) in order to improve hemisphere, and year-round strong westerly and north- understanding of historical limitations to data sets. Although westerly fluxes are found in the midlatitude stormtrack. high-resolution SST data became available as early as 1978, However, while the tropical and subtropical fluxes are quasi- and continuously available from 1982 onward, the accuracy permanent in nature, with relatively little daily variation, the of observations from the advanced very high resolution averaging in Figure 5 masks strong daily variability at the radiometer (AVHRR) was significantly improved by a data- midlatitudes. base that matched these observations to buoy data; this pro- [21] At any time, there are typically three to five major cess of cross-checking began in 1985 [Smith et al., 1996]. conduits in each hemisphere, each of which transports large The Defense Meteorological Satellite Program’s (DMSP) amounts of water vapor in narrow streams from the tropics first SSM/I instrument became operational in July 1987 [e.g., to the higher latitudes. Newell et al. [1992] termed these Robinson, 2004, and references therein]. This sensor brought conduits “atmospheric rivers” (ARs), because they transport about several improvements to the reliability of the variables water at volumetric flow rates similar to those of the world’s used in the data sets of both evaporation and precipitation. largest rivers. These structures account for most of the long- For evaporation, for example, SSM/I was the first satellite to distance transport of water vapor and contain 95% of the provide estimates of sea surface roughness, and consequently meridional flux of water vapor at latitude 35 [Zhu and of wind speed [Goodberlet et al., 1990], as well as specific Newell, 1998; Ralph et al., 2004]. In contrast to terrestrial surface humidity and precipitation from estimates of TCWV rivers, however, these conceptual ARs change course every [Chou et al., 2003]. day with shifting synoptic patterns, and it is only their net [24] For tropical regions, it is possible to use infrared effect (moisture transport from the (sub)tropics east- measurements from geostationary satellites to provide esti- northeastward to the high midlatitudes) that can be seen in mates of precipitation because a strong correlation exists Figure 5. The term “atmospheric river” is not universally between the height and temperature of the top of tropical accepted, and others have suggested different names such as clouds and precipitation [e.g., Adler et al., 2003, and refer- “moisture conveyor belt” [Bao et al., 2006] or “tropical ences therein]. However, such observations are spatially moisture export flow” [Knippertz and Wernli, 2010]. In their limited to the area over which the satellite is positioned, and
8of41 RG4003 GIMENO ET AL.: SOURCES OF CONTINENTAL PRECIPITATION RG4003
Figure 6. Daily integrated total column of water vapor showing the AR that affected the UK on 19 November 2009. Data: ERA-Interim. uncertainties increase toward higher latitudes [Schanze et al., [28] The basic theories used by the scientific community 2010]. to estimate evapotranspiration are the Monin-Obukhov [25] Hence, various “merged” satellite and gauge analyses similarity theory, the Bowen ratio method, and the Penman- have been made in an attempt to maximize the benefits of Monteith equation. The advantages and disadvantages of the using both satellite and gauge measurements of precipitation six major methods of measuring evapotranspiration (EC, BR, [e.g., Adler et al., 2003; Xie and Arkin, 1997]. Uncertainties weighable lysimeters, scintillometer, surface water balance, associated with measurements of precipitation collected by gauge with careful maintenance should be less than about 10% for liquid precipitation but can be much larger for sat- TABLE 2. Date of First Continuous Availability of Different a ellite retrievals and for solid forms of precipitation. Data Sources [26] The incompleteness of records reduces the accuracy of estimates of freshwater discharge from the land to oceans Data Variable Source Available [Di Baldassarre and Montanari, 2009; Legates et al., 2005]. E All In situ and NWP 1948 c Tsea AVHRR 1985 Furthermore, nonriverine flows that connect to coastal sur- AMSR-Eb 2002 face waters, such as from submarine groundwater discharge Uair SSM/I 1987 or seawater inflow, have not been adequately observed QuikSCAT 1999 [Michael et al., 2005]. Consequently, few global analyses of Tair In situ/NWP only 1948 Qair SSM/I 1987 riverine outflow have been made to quantify the freshwater AIRSb 1999 discharge from the land to the oceans [Dai and Trenberth, PPtotal In situ and NWP 1948 OPI 1979 2002; Wang and Dickinson, 2012]. GPI 1986 [27] The Gravity and Climate Experiment (GRACE) sat- SSM/I 1987 ellite [Tapley et al., 2004a, 2004b] was launched in 2002 TOVS 1987 b and allows estimates to be made of the change in terrestrial TRMM-TMI 1997 water storage on a regional and global scale. The spatial low- aIn situ measurements prior to 1948 are not considered. Only commonly resolution ( 200 km) gravimetric data are adequate for used satellite missions that have enhanced the data quality significantly are listed. New sources are only listed if they provide a potential significant studies of large basins, but it does not provide reliable esti- advantage in the future. mates for medium-scale river basins [Werth and Güntner, bThese data sources are not commonly used in order to preserve data homogeneity. 2010]. GRACE also has problems with near-coastal rivers c “ ” Even though AVHRR was first launched in 1978 and was fully and watersheds because of coastal leakage. operational from 1981 onward, sufficient buoy data to constrain the data only became available after 1985. From Schanze et al. [2010].
9of41 RG4003 GIMENO ET AL.: SOURCES OF CONTINENTAL PRECIPITATION RG4003 and atmospheric water balance) were summarized in Table 1 relative to the mean annual evapotranspiration are in tran- [Wang and Dickinson, 2012]. While surface- and satellite- sition zones between dry and humid regions and monsoon based measurement systems can provide accurate estimates regions [Vinukollu et al., 2011]. of the diurnal, daily, and annual variability of evapotranspi- [31] A key source of uncertainty in the reanalysis data is ration, their reliability for longer timescales is poor. The the possible violation of the freshwater cycle, because the surface water budget method can provide a reasonable esti- underlying prediction models are generally forward inte- mate of global mean evapotranspiration, but its regional grating [Wunsch and Heimbach, 2007]. The moisture budget distribution is still rather uncertain. Current land surface is generally not closed in the reanalyses owing to the analysis models provide widely differing values for the ratio of tran- increment that arises from errors in the state variable fields spiration by vegetation to total evapotranspiration. This and observational uncertainties and also a very small term source of uncertainty therefore limits the ability of models to that represents a negative filling to ensure that values of q provide the sensitivities of evapotranspiration to precipitation and w are positive definite [Trenberth et al., 2011]. deficits and changes in land cover. Recent evaluations of Although the reanalyses produce quite good results for pre- global evapotranspiration using different methodologies cipitation over land, over the ocean E, P, and E-P based on indicate great uncertainty across the data sets, of the order of model output are not stable [Trenberth et al., 2011]. The 50% of the global annual mean value [Vinukollu et al., 2011]. poorer representation of coastlines and orography may be a [29] Advances in computer technology have allowed the source of uncertainty in low-resolution reanalyses. When use of computational fluid dynamics and numerical weather coastal ranges are too smooth, the onshore advection of prediction for large data assimilation reanalysis projects, moisture can be excessive [Trenberth et al., 2011]. such as the NCEP Global Reanalysis Project 1 [Kistler [32] Most reanalysis models, with the exception of et al., 2001], hereafter NCEP-1, available from 1948 to the MERRA, predict water cycling (P and E) that is too intense present, the NCEP Global Reanalysis Project 2 [Kanamitsu over the ocean, although ocean-to-land transports are very et al., 2002], hereafter NCEP-2, which uses only satellite close to their observed values [Trenberth et al., 2011]. The data for the whole of the period of analysis (1979–present), results from all the available reanalyses for the main atmo- the Modern Era Retrospective-Analysis for Research and spheric components of the hydrological cycle are given in Applications [Bosilovich et al., 2006], hereafter MERRA Figure 7 for 2002–2008 [from Trenberth et al., 2011]. All (1979–present), the European Centre for Medium-Range P ocean estimates are high relative to the estimate of GPCP. Weather Forecasts (ECMWF) Re-Analysis 40 [Uppala et al., Apart from MERRA, E ocean estimates from reanalyses are 2005], hereafter ERA-40, which is available for 1957–2002, also high when compared with the reference values used and the ERA-Interim data set [Dee et al., 2011]. herein. [30] However, the homogeneity of any reanalysis model [33] Recent reanalyses make use of either a four- is strongly dependent on the homogeneity of the input data dimensional system of data assimilation [e.g., Simmons et al., [e.g., Schanze et al., 2010; Trenberth et al., 2011], which can 2010] or an incremental analysis update technique [Bloom be demonstrated by the climatological discontinuities due to et al., 1996], both of which allow the analyzed fields to the introduction of satellite data in the NCEP-1 reanalysis evolve smoothly in time, rather than in sudden jumps at times [Sturaro, 2003], as well as in ERA-40 [Sterl, 2004]. Schanze of analyses, which reduces the spin-up problem in simula- et al. [2010] evaluated the current quantification of the tions of the hydrological cycle [Trenberth et al., 2011]. oceanic freshwater cycle using new observations from sat- [34] As part of the World Climate Research Program’s ellite data and reanalysis models for evaporation and pre- (WCRP) Global Energy and Water-Cycle Experiment cipitation over the oceans. They found discontinuities in the (GEWEX) Continental-scale International Project (GCIP), a year 1987 for all data sets, which they attributed to the launch preliminary water and energy budget synthesis (WEBS) was of the SSM/I microwave imaging satellite. There are con- developed by Roads et al. [2003] for the period 1996–1999 siderable variations in the precipitation obtained from rea- from the “best available” observations and models. Accord- nalyses that incorporate moisture from satellite observations; ing to these authors, observations cannot adequately char- such variations are a reflection of the changes in the obser- acterize budgets because too many of the fundamental vational system used [Trenberth et al., 2011]. These changes processes are missing. Models that properly represent the also affect the quality of the satellite-derived evapotranspi- many complex atmospheric and near-surface interactions are ration data set [Vinukollu et al., 2011], as well as the esti- also required. mation of evaporation via reanalysis models, because this is estimated using bulk flux formulas. The surface variables 3. SOURCES AND SINKS OF ATMOSPHERIC required for a bulk flux formulation must be estimated from MOISTURE finite values of moisture and temperature for a given layer, 3.1. Methods Used to Establish Source-Receptor which can change over time as satellite instruments change Relationships [Schlosser and Houser, 2007]. In the high-latitude extra- [35] Three principal methods are available for identifying tropics, where remote sensing is much less reliable, studies the source and sink regions of atmospheric moisture, namely have shown that the oceanic satellite estimates of precipita- analytical or box models, numerical water vapor tracers, and tion are less accurate when compared with reanalysis data physical water vapor tracers (isotopes). [e.g., Sapiano et al., 2008]. The greatest uncertainties
10 of 41 RG4003 GIMENO ET AL.: SOURCES OF CONTINENTAL PRECIPITATION RG4003
Figure 7. Estimated values of the observed hydrological cycle using eight reanalyses for 2002–2008, with the exception for ERA-40, which starts from 1990 (color coded as given at the bottom of the figure). For the ocean-to-land water vapor transport, the three estimates given for each are (1) the actual transport estimated from the moisture budget (based on analyzed winds and moisture), (2) E-P from the ocean, and (3) P-E from the land, which should be identical. Units: 1000 km3 yr 1. Adapted from Trenberth et al. [2011].
3.1.1. Analytical or Box Models estimate of the recycling that takes place within a region (see [36] The underlying motivation for the development of Burde and Zangvil [2001a] for a derivation of the model). analytical models to show the source and sink regions of After Budyko’s initial conceptualization, a number of atmospheric moisture has historically been an understanding authors have developed models to expand and improve the of how changes in the surface hydrology of a region, due to quantification of precipitation recycling. The initial 1-D anthropogenic influences or natural variability, are likely to approach was later extended to two dimensions [Brubaker et modify the climate through changes in the water cycle al., 1993; Eltahir and Bras, 1996; Burde and Zangvil, [Eltahir and Bras, 1996; Brubaker et al., 1993]. 2001a, 2001b; Savenije, 1995]; however, all these models [37] The earliest quantitative theory and analytical models continued to work on monthly or longer timescales, and of source-sink regions focused on the contribution of hence the first term in equation (7) could be neglected. evapotranspiration to local precipitation, or precipitation Dominguez et al. [2006] later developed the “Dynamic recycling. All analytical models can be derived from the Recycling Model (DRM)” in which the assumption of neg- equation of the vertically integrated balance of water vapor ligible moisture storage was relaxed, and the model could (following the review of Burde and Zangvil [2001a]): then be used at timescales shorter than a month. In the DRM, equation (7) is solved in a Lagrangian framework, and the ∂ðÞw ∂ðÞwu ∂ðÞwv þ þ ¼ E P; ð7Þ local recycling ratio R (the amount of precipitation for a ∂t ∂x ∂y particular cell that originates as evapotranspiration within where w is the amount of water vapor contained in a column the selected region) is Z of air of unit base area, u is the vertically integrated zonal t E water vapor flux divided by w (this is equivalent to a water R ¼ 1 exp dt’ ; ð8Þ 0 W vapor-weighted zonal wind), v is the water vapor weighted meridional wind, E is evaporation, and P is precipitation. where E is evapotranspiration and W is precipitable water, The equation can be used separately for moisture entering calculated at different times t, following the trajectory of the the region from the outside (advection) and for moisture parcel. When applied to monthly timescales, the DRM esti- originating within it (recycling). Budyko and Drozdov mates very similar spatial and temporal variability of recy- [1953] and later (in English) Budyko [1974] developed a cling to the Brubaker et al. [1993] and Eltahir and Bras model by assuming the following: (1) a negligible change in [1996] models, but the estimates are slightly higher. In storage of atmospheric water, (2) a one-dimensional (1-D) addition, the DRM can be used to calculate particular source estimation of recycling, and (3) a well-mixed atmosphere. and sink regions of precipitation [Dominguez et al., 2008], Considering the basic equation of the conservation of mass, making it more versatile than the traditional bulk models. At assumptions (1) and (2) imply that the first and third terms in about the same time, Burde et al. [2006] relaxed the equation (7) may be neglected. This is then a simple 1-D assumption of a well-mixed atmosphere by accounting for
11 of 41 RG4003 GIMENO ET AL.: SOURCES OF CONTINENTAL PRECIPITATION RG4003 the “fast” recycling that takes place when the precipitation prognostic equation for any given water vapor tracer that originates from evapotranspiration does not mix with follows: advected moisture. This model can be used in regions where ∂ ∂ ∂ qT qT qT the ratios of recycled to total precipitation, and precipitable ¼ D3 ðÞþqTV þ ðÞEsurf þ fc ∂t ∂t T ∂t water, are known. turb cond ∂ ∂ [38] The foregoing analytical models have generally been þ qT þ qT ; ð9Þ fR ∂ fRAS ∂ applied to specific regions at the subcontinental scale. The t revap t RAS estimates of recycling are a function of the size of the area under consideration, where the recycling increases with the which indicates that the changes in the water vapor tracer are area considered. However, there is a strong logarithmic affected by advection by winds, turbulence including con- relationship between recycling ratio and area for different vection (turb), evaporation in the source region of the tracer, regions of the world [Brubaker et al., 2001; Dominguez condensation (cond), rain evaporation (revap) and redistri- et al., 2006; Dirmeyer and Brubaker, 2007], which bution by convection (RAS), and the f terms are propor- allowed Dirmeyer and Brubaker [2007] to scale recycling tionality relationships. One potential limitation of the to a common area and produce a meaningful global gridded numerical WVT approach is that the results depend on how analysis of the recycling ratio. realistically the numerical model can simulate all the rele- [39] An alternative approach is via the evaluation of the vant processes. percentage of precipitation falling in a region that originates [41] During recent years, the use of Lagrangian methods as continental evapotranspiration or “continental precipita- has become popular for diagnosing the transport of moisture tion recycling ratio” (as opposed to “local” evapotranspira- and, in particular, for determining the origin of moisture that tion). To do this, van der Ent et al. [2010] formulated a precipitates in particular regions. At first, simple back tra- variation of the traditional analytical models using a jectories from areas of precipitation were used to infer the numerical solution of the same underlying equation of origins of air masses [e.g., D’Abreton and Tyson, 1995]. atmospheric moisture balance (equation (7)). This formula- Precipitation rates were calculated from the decrease of tion allows the estimation of the percentage precipitation of specific humidity along trajectories [Wernli, 1997] and then terrestrial origin at the global scale. Using this numerical used to diagnose the origin of the moisture for heavy pre- approach, Keys et al. [2012] were able to delineate “pre- cipitation events [Massacand et al., 1998]. Dirmeyer and cipitation sheds,” or evaporation source areas that contribute Brubaker [1999] and Brubaker et al. [2001] combined moisture to precipitation downwind. Unlike terrestrial large sets of back trajectories using gridded information on watersheds, precipitation sheds are variable in space and evaporation and precipitation rates (generally from reanaly- time. The concept of a precipitation shed is useful for sis data), accounting for uptake and loss of moisture as the understanding how precipitation in regions depends on trajectories pass over these sources and sinks. In this upwind surface hydrological conditions. method, described in Dirmeyer and Brubaker [1999], back 3.1.2. Numerical Water Vapor Tracers trajectories are computed from each grid square at which [40] The second method of studying source-sink regions precipitation has occurred. Parcels are launched backward in makes use of numerical water vapor “tagged” tracers time at a rate proportional to the precipitation, from a vertical (WVT), which is also known as a water vapor “tagging” location that is determined probabilistically depending on approach. We can divide these methods into Eulerian and the moisture at that level. As a parcel (k) is tracked backward Lagrangian. In the Lagrangian frame of reference the in time, the fraction of precipitable water (W) of the parcel observer follows an individual fluid parcel as it moves assumed to have been contributed by surface evaporation (E) through space and time. On the other hand, the Eulerian at point (x, y) at each time step (t)is frame of reference focuses on specific locations in the space ExðÞ; y; t R ; ðÞ¼x; y; t ð10Þ through which the fluid flows as time passes. Initially i k W developed by Joussaume et al. [1984] and Koster et al. i
[1986], Eulerian tagging techniques not only yield infor- where Ri,k(x, y) represents the evaporative contribution of mation on recycled precipitation but also account for the surface grid (x, y) to the precipitable water that contributed to specific origin and destination of advected moisture. rainfall in grid box (i) from parcel (k). The total mass con- Numerical tracers are implemented in GCMs and experience tribution of evaporation from grid square (x, y) to precipi- the same processes as atmospheric water. Because they are tation on an area A with a total of n grid squares is then embedded in climate models, numerical WVT models calculated using all k parcels launched from A. incorporate state-of-the-art understanding of how moisture moves and is transformed as it passes through the atmo- Xn Xk Xtf EAðÞ¼x; y Ri;k ðÞx; y; t ; ð11Þ sphere. Bosilovich and Schubert [2002] described the use of i¼1 k¼1 t¼0 numerical WVTs specifically to address the question of recycling. In their study, as in the studies of Koster et al. where tf is the ending time of the longest back trajectory [1986] and Joussaume et al. [1984], passive constituents calculation. in the GCMs are predicted forward in time, in parallel [42] This method allows a detailed budget of moisture with the prognostic water vapor variable of the model. The along the trajectories and provides estimates of precipitation
12 of 41 RG4003 GIMENO ET AL.: SOURCES OF CONTINENTAL PRECIPITATION RG4003 recycling. However, unlike the Eulerian tracer methods, the validated using physical measurements. The heavy stable transport of and changes in water vapor do not depend on the isotopes of hydrogen and oxygen, D (deuterium) and 18Oin detailed physical equations of the underlying reanalysis precipitation and/or water vapor, are ideal measurable para- model. meters because they are an integrated product of both the [43] Subsequently, Stohl and James [2004, 2005] devel- history of an air mass and the specific prevailing meteoro- oped an analog method that accounts for the net loss and/or logical conditions (temperature as well as humidity and wind gain of moisture along trajectories using speed) at the time of condensation [Gat and Carmi, 1970]. The isotopic compositions are usually denoted dD and d18O dq ðÞe p ¼ m ; ð12Þ and expressed in parts per thousand (‰) relative to the k dt standard mean ocean water (SMOW) composition. Because 16 16 16 where (e-p)k are the rates of increase and decrease of moisture of differences in mass, mixtures of H2 O/HD O and H2 O/ 18 along the trajectory of each particle and (q) is the specific H2 O have different chemical and physical properties. humidity taken from the meteorological (e.g., reanalysis) Therefore, when the water changes phase, the heavy isotopes 16 18 data, which are also used as input to the Lagrangian model. (HD O and H2 O) become preferentially enriched in the By filling the atmosphere with a large number of computa- liquid rather than the gas phase and in the solid rather than tional air particles, the surface freshwater flux in an area A the liquid phase. This is called isotopic fractionation. Phase can be determined using changes always occur during the circulation of atmospheric X water, and geographical and temporal differences in isotopic K ðÞ ¼ e p ratios therefore emerge in vapor and precipitation. It is E P ¼ k 1 k ; ð13Þ A noteworthy that no fractionation occurs between the water taken up by and transpired from plants because of the fact where a budget is calculated for all K particles that reside that isotopic fractionation actually occurs against leaf water. above A. Thus, the surface freshwater flux E-P can be [47] By adding the isotopic processes in the analytical and accounted for, using information on the trajectories of the numerical models and by comparing modeled and measured particles. Net loss or gain of moisture can be identified both isotopic composition in precipitation and/or water vapor, along individual particle trajectories as well as on a regular one can directly validate the model’s transport processes. grid, using only particle information. With this methodology, These types of validation are common, both in studies of the evaporative source and sink regions for a given area can atmospheric vapor cycling during large-scale transport (e.g., be identified and linked using the trajectory information. Yoshimura et al. [2004], where large-scale moisture flux in [44] The method of Stohl and James [2004, 2005] differs major reanalysis products is validated) and for in-cloud from that of Dirmeyer and Brubaker [1999] in a number of processes [e.g., Blossey et al., 2010], where isotopic pro- respects: (1) the trajectory information is obtained from a cesses associated with all microphysical interactions were particle dispersion model [Stohl et al., 1998] and includes incorporated in a cloud-resolving model. Furthermore, sub-grid turbulence [Stohl et al., 2005], and (2) the only recycling due to transpiration in Amazonia was suggested by input to the moisture diagnostics is the change in specific Salati et al. [1979] using evidence of a decrease of isotopic humidity with time, while Dirmeyer and Brubaker [1999] depletion with distance from the coast. This was revisited by use evaporation and precipitable water. Henderson-Sellers et al. [2002] in their investigation of the [45] One disadvantage of the Stohl and James [2004, 2005] deforestation and warming in Amazonia. method is that evaporation and precipitation are not clearly [48] Notice, however, that two additional isotopic tracers separable. Furthermore, the quantity (E-P) is obtained using are not sufficient to constrain all influencing processes. the time derivative of humidity along the particle trajectories. Furthermore, the isotopic fractionations during evaporation In consequence, if the reanalysis data used to drive the model from surface water [Craig and Gordon, 1965; Merlivat and do not properly close the water budget (in fact, the analysis Jouzel, 1979] and from falling droplets in a cloud [Stewart, increment is often the dominant term in the budget), then the 1975], as well as the reevaporation from land and plant method may suffer from considerable inaccuracies. In fact surfaces are often not described accurately by available this last inconvenience is shared with Dirmeyer and parameterizations. Brubaker [1999] method since this is based on calculated [49] Isotopic data related to precipitation have been col- evaporation, which is probably the most uncertain term and it lected since the 1960s. With the worldwide effort led by the also does not close the water budget. Lagrangian methods International Atomic Energy Agency/World Meteorological have been used to study the origin of water that falls during Organization (IAEA/WMO), Dansgaard [1964] suggested a extreme precipitation events [e.g., Stohl et al., 2008; temperature effect, a latitudinal effect, an altitude effect, and Gustafsson et al., 2010]. However, these methods are also an amount effect on isotopic composition. These effects sufficiently computationally efficient to establish the cli- have been repeatedly confirmed by others following differ- matologies of moisture source-receptor relationships [e.g., ent observational studies. Friedman et al. [1992] measured Stohl and James, 2005; Gimeno et al., 2010a]. the isotopic composition of precipitation samples at numer- 3.1.3. Physical Water Vapor Tracers ous sites in southeastern California over a 7 year period, and [46] Although analytical and numerical models are pow- based on seasonally integrated samples, they suggested that erful tools for studying atmospheric recycling, they must be atmospheric circulation is likely to be the leading cause of
13 of 41 RG4003 GIMENO ET AL.: SOURCES OF CONTINENTAL PRECIPITATION RG4003 isotopic variability. Other studies have also shown that the those on satellites [e.g., Schneider et al., 2010]. Recently, isotopic composition of rain in individual storms is closely precise optical analyzers for in situ HDO measurements have tied to a storm’s trajectory [Benson and Klieforth, 1989; become available [e.g., Lee et al., 2006; Welp et al., 2008]. Friedman et al., 2002; Ingraham and Taylor, 1991]. Isotopic The combination of these new measurements from satellites variability among storms also results from local meteoro- and ground truth observations will provide a wealth of logical conditions [Coplen et al., 2008], and much of this information for future studies. variability has to do with dynamical processes during a [52] The isotope-incorporated atmospheric general circu- storm’s evolution in addition to the isotopic variability of the lation models (AGCMs) initiated by Joussaume et al. [1984] vapor source, because of changes in wind speed/direction have recently gained in popularity [e.g., Yoshimura et al., [Fudeyasu et al., 2008; Yoshimura et al., 2008]. 2008; Risi et al., 2010a]. The work of the stable water iso- [50] Stable water isotopes are also a useful tool for parti- tope modeling intercomparison group (SWING) is now into tioning fluxes of evaporation and transpiration at the eco- its second phase, and there are more than ten isotope- system scale and their use has been steadily increasing incorporated AGCMs and a few regional climate models [Moreira et al., 1997; Yakir and Sternberg, 2000; Yepez (RCMs) used for this purpose [Noone and Sturm, 2010]. By et al., 2003; Williams et al., 2004; Yakir and Wang, 1996; combining the recent vapor isotope observations described Wang and Yakir, 2000; Ferretti et al., 2003; Yepez et al., above with AGCM results, Risi et al. [2010b] pointed out the 2007]. Evaporation and transpiration fluxes have distinc- potential of isotopic information to find areas of misrepre- tive isotopic compositions. Evaporated water is significantly sentation of the model in terms of dehydrating processes in lighter than transpired water because when the latter leaves the Sahel region associated with the subsidence of the Hadley the stomata, it remains isotopically closer to that taken up by cell. Similarly, Yoshimura et al. [2011] showed the large- the plant because unfractionated water is continuously being scale agreement between the AGCM and the satellite-based replenished through the stem; in fact when transpiration is at vapor isotopic distributions. They concluded that the isotopic steady state (ISS) there is no isotopic fractionation parameterization for reevaporation from a falling droplet in a and the isotopic composition of transpired vapor can be the convective cloud affected the isotopic composition in the same as that of the stem water [Farquhar and Cernusak, mid-troposphere over the Maritime Continent (Figure 8). 2005]. On the other hand, evaporation from the soil and 3.1.4. Intercomparison of the Source-Receptor wet surfaces is heavily fractionated as lighter isotopes are Methods preferentially transferred to the vapor phase [Craig and [53] The establishment of the source-receptor relationship Gordon, 1965]. may often be best achieved in an integrated manner, using [51] Until recently, observations of the isotopic composi- the results gathered from several of the different methods tion of water vapor were severely lacking because traditional described herein. The use of Eulerian fields provides the isotopic measurement techniques are somewhat complex large-scale characteristics of circulation involved in the (e.g., the cryogenic method). Recent advances in remote transport and together with numerical WVT is constrained sensing of vapor isotopes from satellites, particularly HDO by the input data, which in turn depend on the numerical (heavy water where one proton is replaced by deuterium), models. Lagrangian models may be used to assess the geo- have dramatically increased the availability of observed data. graphical origin of moisture that reaches a region. Physical After Zakharov et al. [2004] first retrieved latitudinal cli- WVTs (isotopes) are very useful for model validation. matology for column vapor HDO using IMG (the Interfero- Table 3 summarizes the main advantages and disadvantages metric Monitor for Greenhouse gases sensor) on ADEOS of each methodology. To illustrate the main points, two (Advanced Earth Observing Satellite), Worden et al. [2006] continental regions were chosen in order to compare results then retrieved data on low-level atmospheric vapor HDO. obtained using the different methods, namely Spain and Over tropical regions at fine temporal and spatial resolutions the Orinoco River basin. The first region is located in the using TES (Tropospheric Emission Spectrometer) on the extratropics, with the extratropical storm track being the satellite Aura, Payne et al. [2007] retrieved monthly data on principal mechanism of precipitation [Trigo et al., 1999]; the global distribution of upper troposphere and stratosphere and the second is located in the tropics, where the dis- vapor HDO using MIPAS (the Michelson Interferometer for placement of the ITCZ is the dominant factor in the precip- Passive Atmospheric Sounding) on Envisat (environmental itation regime [Poveda et al., 2006]. Information on the satellite), and Frankenberg et al. [2009] measured the atmo- sources of moisture derived from the different methods (box spheric column vapor deuterium ratio using SCIAMACHY models, Eulerian fields, numerical WVT and isotopes) for (Scanning Imaging Absorption Spectrometer for Atmospheric these two regions is shown in Figures 9 and 10. This allows Chartography), also on Envisat. Although some limitations us to contrast the detail provided by each type of method and remain in terms of spatial and temporal coverage, resolution, to show the complementary nature of the information. It is precision, and accuracy, the resulting maps have improved also of some interest to note the differences between a region the general understanding of the distribution of isotopes and where a vast number of specific studies and observational the physical processes that trigger the isotopic distributions. networks is available (Spain) and a region where observa- It is also worth mentioning that remote sensing has been tions and detailed analyses have historically been few widely used with several ground-based Fourier transform (Orinoco River basin). spectroscopy instruments, which are essentially the same as
14 of 41 RG4003 GIMENO ET AL.: SOURCES OF CONTINENTAL PRECIPITATION RG4003
Figure 8. (a) Mean climatology of dD in midtropospheric water vapor (800 to 500 hPa pressure) for TES; (b) sensitivity simulation (E10) with an isotope-incorporated AGCM (IsoGSM), in which isotopic fractionation with reevaporation from falling droplets in convective clouds is more suppressed; (c) difference between the satellite measurements and model simulation. The global-scale biases in TES are arbitrarily corrected by +20‰ (indicated by “TES + 20”). Adapted from Yoshimura et al. [2011].
15 of 41 RG4003 GIMENO ET AL.: SOURCES OF CONTINENTAL PRECIPITATION RG4003
TABLE 3. Summary of the Main Strengths and Weaknesses of Analytical Box Models and Physical and Numerical (Eulerian and Lagrangian) Water Vapor Tracer Method
Type Strength Weakness References (Nonexhaustive) Analytical Box Models Simple as few parameters are required Neglects in-boundary processes; Budyko [1974]; and they consider grid based some are based on the well Brubaker et al. [1993]; spatial variability. mixed assumption Eltahir and Bras [1994]; (the local source of water Burde and Zangvil [2001a, is well mixed with all 2001b]; Dominguez et al. [2006]. other sources of water in the whole vertical column); most are only valid for monthly or longer timescales. Physical Water Vapor Tracers Simplicity; global coverage; Sensitivity of the isotopic signal; Gat and Carmi [1970]; include vertical processes; calculation time; availability Salati et al. [1979]; reanalysis input data of data for validation; Rozanski et al. [1982]; (high spatiotemporal resolution); does not account for Coplen et al. [2008]. enable the combination of GCMs convection and rainwater and Lagrangian Rayleigh models. evaporation/equilibration. Numerical Water Eulerian Detailed atmospheric processes; Dependent on the model bias; Benton and Estoque [1954]; Vapor Tracers realistic moisture circulation. global forcing is required; Starr and Peixoto [1958]; poor representation of Peixoto and Oort [1982]; short-timescale hydrological Joussaume et al. [1984]; cycle parameters; Koster et al. [1986]; does not include the remote Bosilovich and Schubert [2002]. sources of water for a region. Lagrangian High spatial resolution moisture Sensitivity of moisture flux D’Abreton and Tyson [1995]; sources diagnostics; computations to increases Wernli [1997]; quantitative interpretation in data noise for shorter time Massacand et al. [1998]; of moisture origin allowed; periods or smaller regions; Dirmeyer and Brubaker [1999]; not limited by a specific RCM simple method does not provide Brubaker et al. [2001]; domain and spin-up; a diagnostic of surface fluxes Dirmeyer and Brubaker [2006]; establishment of source-receptor of moisture; surface fluxes Stohl and James [2004, 2005]. relationship can be easily under (over) estimation if assessed because budgets can dry (cold) air masses tracking be traced along suitably as the budget is not closed; defined trajectory ensembles; evaporation rates are based net freshwater flux can be on calculations rather than tracked from a region both observations in some methods; forward and back ward in time; evaporation and precipitation realistic tracks of air parcels; are not clearly separable computationally efficient compared (in some methods); movement to performing multiyear GCM and extraction of water simulations or reanalyses; does not depend on the more information provided physical tendencies included than a purely Eulerian description in the reanalysis data. of velocity fields; parallel use of information from Eulerian tagging methods allowed.
[54] The analysis was carried out for the 5 year period surrounding water bodies and northern Africa is advected from 2000 to 2004. Using ERA-Interim vertically integrated into continental Spain, as shown by the vectors of moisture water vapor fluxes and vertically integrated moisture flux flux. Regions of strong evaporation are shown in yellowish divergence with a horizontal resolution of 0.5 degrees, a shades, while moisture sinks are shown in bluish colors, as simple box model method was applied within the borders of the regions where precipitation is found to occur. This type Spain to identify the origins of moisture from the moisture of method is the most widely used in the literature for several flux through the borders. Figure 9a shows that the main regions of the globe because of the simplicity and avail- result is moisture inflow from the lateral boundaries, from ability of the analysis data sets. Specific information on the the Mediterranean Sea to the east and from the North moisture related to precipitation over a determined region is Atlantic to the west. The box model allows the identification not immediately available from these fields. 18 of the moisture inflow and outflow, and its approximation is [55] “Long-term” d O values from Global Network of good, but it lacks information on the physical processes Isotopes in Precipitation (GNIP) stations over Spain are between the boundaries and may not be suitable for ana- shown in Figure 9c; the gradients of d18O between western lyzing relatively small regions. Figure 9b shows the Eulerian coastal and inner Iberian Peninsula are in good agreement fluxes, using ERA-Interim data from 2000 to 2004 on a 0.5 with the westerly circulation regime shown in Figure 9b, horizontal grid, in which the large-scale characteristics of the which supports the result from the Eulerian fluxes that the transport of moisture may be seen. Moisture from the North Atlantic is a major source of moisture for precipitation
16 of 41 RG4003 GIMENO ET AL.: SOURCES OF CONTINENTAL PRECIPITATION RG4003
Figure 9. Comparison among the results obtained using different methods for the climatological mean pattern for 2000–2004 for Spain: (a) simple box model showing the moisture flux across the segments of zonal and meridional regional boundaries; (b) typical Eulerian field method using vertically integrated water vapor flux (shaded) and moisture flux vectors (black arrows); (c) long-term weighted delta18Oin precipitation in the GNIP stations; (d) identification of the sources of moisture using ten-day integrated net freshwater flux from FLEXPART backward trajectories (shaded contours) and from quasi-isentropic back trajectory analysis of atmospheric water vapor (solid lines) from Dirmeyer and Brubaker [2006]; see the atlas at http://www.iges.org/wcr. Data, Figures 9a and 9b: ERA-Interim 0.5 resolution. over Spain. The gradient also shows the influence of the colors in Figure 9d) or where they lose moisture (sinks, Mediterranean, again in agreement with the results from the bluish colors). The difference, for the moisture sources of the box model and the Eulerian fluxes. Finally, Figure 9d shows Iberian Peninsula, is evident in an important part of the storm the results for two trajectory methods: the contour lines track area (latitudes higher than 30 in the Mid-Atlantic), show results obtained using quasi-isentropic back trajecto- which is considered to be a moisture source in the quasi- ries [Dirmeyer and Brubaker, 2007; see the atlas at http:// isentropic approach but not in the Lagrangian FLEXPART www.iges.org/wcr], and the shaded colors show the results model. In the latter method, losses of moisture in these obtained using the Lagrangian FLEXPART model method regions are much higher than uptakes; it is not a “true”source of Stohl and James [2004], which accounts for the integrated region for the Iberian Peninsula. net freshwater flux over 10 day periods. Both methodologies [56] The comparison for the Orinoco River basin is shown identify the patterns of the origins of moist air, but the quasi- in Figure 10. Using a simple box model (Figure 10a), the isentropic approach cannot provide information on the importance of moisture inflow from the tropical Atlantic is “history” of the moisture variations along the trajectory, highlighted, as is other inflow further inland. For the Orinoco whereas the Lagrangian FLEXPART model method is able River basin, the role of the tropical Atlantic as a principal to show those areas where the particles along the trajectory source of moisture is well supported for the known circula- gain moisture (evaporative sources of moisture, reddish tion in the region. However, due to the proximity of the
17 of 41 RG4003 GIMENO ET AL.: SOURCES OF CONTINENTAL PRECIPITATION RG4003
Figure 10. As Figure 9 but for the Orinoco River basin.
Amazon region, further detail is required in order to consider [57] Table 4 summarizes the results obtained with the processes associated with recycling or even transport to the various source-receptor diagnostics for two regions that have Amazon, which may exert an influence on the moisture pat- been studied in some detail, namely the Mississippi River terns over the Orinoco River basin. From the Eulerian fluxes basin and the Sahel region, again suggesting that the dif- shown in Figure 10b, the fluxes into the Orinoco basin from ferent methods provide complementary information. the tropical Atlantic, as well as the importance of the inland 3.2. Global Source and Sink Regions of Moisture fluxes, which connect the Orinoco and the Amazon basins [58] The results of the last 20 years of work related to can be noted. Isotopic data for the Orinoco basin is available sources and sinks of precipitation using the methods for a single station (Figure 10c). The comparison between the described above provide us with an understanding of the quasi-isentropic trajectories and the Lagrangian FLEXPART ways in which global evapotranspiration contributes to pre- model shows the marked differences between the two methods cipitation. We will first summarize the results for precipita- (Figure 10d). The identification of the origin of moisture in the tion of oceanic origin and then those for precipitation of first approach considers a broad picture of the source because terrestrial origin. the presence of the ITCZ is lacking. In the second case the 3.2.1. Oceanic Sources presence of the ITCZ is shown in some detail, which is par- [59] The principal oceanic sources of atmospheric moisture ticularly important when studying climate in the tropics. The are summarized in Figure 11 (right). These areas were main difference between results from the method based on defined by Gimeno et al. [2011] using the threshold of 750 quasi-isentropic trajectories [Dirmeyer and Brubaker, 2007] mm yr 1 for the climatological annual vertically integrated and the method based on the Lagrangian FLEXPART model moisture flux divergence in the ERA40 reanalysis data set for [Stohl and James, 2004] is due to the own objective of each the period 1958–2001 shown in Figure 11 (left) (only two method: the former diagnoses E, whereas the latter diagnoses sources of moisture were defined using the physical E-P.
18 of 41 RG4003 GIMENO ET AL.: SOURCES OF CONTINENTAL PRECIPITATION RG4003
TABLE 4. Summary of the Key Results Obtained From Selected Papers for the Mississippi River Basin and the Sahel Region Using Isotopes and Eulerian and Lagrangian Methodologies to Study the Source-Receptor Relationships
Method Key Result Reference Mississippi Isotopes High evaporation in the lower Mississippi; Kendall and Coplen [2001]; locally derived groundwater is a source for nearby streams; Vachon et al. [2010]. latitudinal gradients in the Mississippi River valley are steeper during cold months. Eulerian Flood events have a strong link with local surface evaporation Trenberth and Guillemot [1996]; as recycling decreases while evaporation from Helfand and Schubert [1995]. the IAS is increased; the inflow of moisture from the south is dominated by the LLJ. Lagrangian Precipitation and recycling are correlated with evaporation Dirmeyer and Brubaker [1999]; at an interannual scale; evaporation is related to the moist Bosilovich and Chern [2006]; and shallow PBL that provides moisture for convection; Brubaker et al. [2001]; recycling is partly correlated with warm SSTs in the Stohl and James [2005]; tropical Pacific Ocean; recycling and evaporation Gimeno et al. [2010a]. from the ocean are the dominant sources of moisture during spring whereas recycling is the dominant source during summer.
Sahel Isotopes Recycling is the major source of moisture for precipitation; Bowen and Revenaugh [2003]; precipitation decreases at the onset of the monsoon Bowen [2009]; as the ITCZ shifts northward from the Guinean Risi et al. [2010a, 2008]. coast to the Sahel. Eulerian The Gulf of Guinea and its northern belt are a source Cadet and Nnoli [1987]; of water vapor transported northward; Druyan and Koster [1989]; moisture convergence and divergence patterns Bielli and Roca [2010]; over northern Africa influence rainfall over Gong and Eltahir [1996]; sub-Sahara more than evaporation or moisture Fontaine et al. [2003]; advection over/from over the adjacent oceans; Pu and Cook [2011]. recycling is a main source of precipitation over the Sahel in the “rainy” season; south of the Sahel, correlation between precipitation and evaporation is negative and large scale; evaporation over the Sahel peaks 1–3 days after precipitation, maximum contribution from small-scale processes occurs during the first day; over the western Africa two-thirds of rainfall at the seasonal scale being advected from the tropical Atlantic and central Africa, the remainder is recycling; moisture advected into WAM region originates in the Mediterranean Sea and central Africa; westerly moisture flux variability related to variations in the jet trigger variations in the content of low-level moisture, modulating atmospheric stability. Lagrangian Recycling was identified as the major source of moisture; Nieto et al. [2006]; important contributions from a band along the Dirmeyer and Brubaker [2006, 2007]. North Atlantic from the Sahel latitudes to the Iberian Peninsula coast; the Mediterranean Sea and the Red Sea are other important sources (note that these sources in some Lagrangian methods are likely erroneously large); there is a strong moisture uptake over the tropical South Atlantic following the fifth day of transport, including the Guinea Gulf, during summer; the Indian Ocean does not seem to be an important source, although it could have a minor influence during summer.
boundaries of oceanic basins, namely the Mediterranean and obtained via forward tracking from the source areas using the the Red Sea). Though the data and periods used in Figures 5 Lagrangian method of Stohl and James [2004] (Figure 11, and 11 are different (multiple satellite observations for 1999– right, for the period 1980–2000). The productivity of the 2008 [Xie et al., 2008] and ERA40, for 1958–2001), the major oceanic sources of moisture is not evenly distributed, regions with higher vertically integrated moisture fluxes and some specific oceanic sources are responsible for more (reddish colors) occur over the same oceanic areas. The continental precipitation than others [Gimeno et al., 2010a]. continental receptor regions of the evaporated moisture were
19 of 41 RG4003 IEOE L:SUCSO OTNNA PRECIPITATION CONTINENTAL OF SOURCES AL.: ET GIMENO 0o 41 of 20